A Text-Syntax Fusion Coreference Resolution Framework for Conversational System

Coreference resolution is a crucial task in conversational systems. However, most existing studies focus on only modeling the text information. Even though a few methods attempt to utilize syntactic information, but they neglect the syntactic relationship between different words. In addition, they incorporate textual and syntactic information by direct concatenation. To address these problems, this paper presents a Text-Syntax Fusion based end-to-end Framework (TSFF). To capture multiple syntactic relationships between words, TSFF adds independent nodes for words with the same syntactic relationship to represent their explicit relationship with each other, simplifying the modeling process of the graph representation. To discover the correlation between different features, TSFF utilizes co-attention to fuse syntactic and textual features. Extensive experiments on two real-world datasets demonstrate the effectiveness of our proposed framework as well as the utility of considering syntactic relationship between different words and the co-attention fusion strategy.